Abstract

Parkinson’s Disease (PD) is a neuro-degenerative disorder, affecting the mobility of persons. The symptoms of PD include trembling, tight muscles, and unsteady walking motions. PD has been classified in various studies in the past, although in this work an attempt is made to differentiate between patients suffering from PD and the healthy persons by concentrating on the specific aspects of gait rhythms. The experiment is performed on the dataset with 15 PD patients and 16 healthy control individuals. Eight statistical features are extracted from the thirteen timeseries gait data. As feature selection plays a crucial role in improving model’s performance, an optimal subset of features is obtained by calculating Mutual Information Gain (MIG) and by Recursive Feature Elimination (RFE). The top 10% features are chosen from the extracted statistical features and they are classified separately. In the next step, the features obtained by both the techniques are then concatenated together and further classified using Machine Learning classifiers to enhance the model’s performance. In the current work, we have proposed a hybrid MIG-RFE feature selection approach for classification of PD from healthy people using the gait data. When evaluated using 10-fold cross validation technique, the proposed MIG-RFE feature selection approach provided the maximum classification accuracy of 96.82% by Naïve Bayes classifier with only fourteen features. The experimental analysis shows that the obtained results are better than some of the state of art methods.